Echo signals processing with neural network in bat-like sonars based on PVDF

Abstract

The sophisticated sonar system of bats is capable of extracting a complete set of information in order to locate and characterize the prey. Curved piezopolymer transducers were already used in the same frequency range to generate ultrasonic waves in air. In this article we investigate the possibility to process echo-signals by using a neural network, similarly to the biological model. We analyse frequency modulated signals, which are only a part of the most complex bat echo signal, in order to measure the target distance through the time of flight evaluation. Chirps are first amplified with a low noise CMOS amplifier, than are properly filtered and rectified in order to obtain a pulse time sequence. Finally the pulse signal is processed by the first level of a neural network to recognize the right pulse time sequence and evaluate the time of flight. Simulated results carried out by using both Matlab and Spice programs, will be presented with reference to piezo-polymer transducers application.